Sampled-data control of nonlinear systems:

an overview of recent results

Dragan Nešić1 and Andrew R. Teel2

1
Department of Electrical and Electronic Engineering, The University of
Melbourne, Parkville, 3010, Victoria, Australia
2
CCEC, Electrical and Computer Engineering Department,
University of California, Santa Barbara, CA, 93106-9560

Abstract. Some recent results on design of controllers for nonlinear sampled-data

systems are surveyed.

1 Introduction

The prevalence of digitally implemented controllers and the fact that the
plant dynamics and/or the controller are often nonlinear strongly motivate
the area of nonlinear sampled-data control systems. When designing a con-
troller for a sampled-data nonlinear plant, one is usually faced with an im-
portant intrinsic difficulty: the exact discrete-time/sampled-data model of
the plant can not be found.
The absence of a good model of a sampled-data nonlinear plant for a
digital controller design has lead to several methods that use different types
of approximate models for this purpose. We single out two such methods
that attracted lots of attention in the sampled-data control literature and
that consist of the following steps:

Method 1: Method 2:

continuous-time plant model continuous-time plant model

⇓ ⇓
continuous-time controller discretize plant model
⇓ ⇓
discretize controller discrete-time controller
⇓ ⇓
implement the controller implement the controller

Hence, in Method 1 (this method is often referred to as the controller

emulation design) one first designs a continuous-time controller based on a
continuous-time plant model. At this step the sampling is completely ignored.
Then, the obtained continuous-time controller is discretized and implemented
using a sampler and hold device. On the other hand, in Method 2 one first
2 D. Nešić and A.R. Teel

finds a discrete-time model of the plant, then designs a discrete-time con-

troller based on this discrete-time model and finally implements the designed
discrete-time controller using a sampler and hold device.
In principle, Method 2 is more straightforward for linear systems than
for nonlinear systems. Indeed, for linear systems we can write down an ex-
plicit, exact discrete-time model while typically for nonlinear systems we
cannot. Moreover, the exact discrete-time model of a linear system is linear
while the exact discrete-time model for a sampled-data nonlinear system does
not usually preserve important structures of the underlying continuous-time
nonlinear system, like affine controls for example. Consequently, for nonlinear
systems it is unusual to assume knowledge of the exact discrete-time model
of the plant whereas this assumption is made in most of classical linear lit-
erature. However, often due to our inability to exactly compute the matrix
exponential that generates the exact discrete-time model of a linear plant, we
use its approximation and hence we actually use an approximate discrete-time
plant model for controller design most of the time, even for linear systems.
Consequently, we will always assume in the sequel that in Method 2 we use
an approximate discrete-time model of the plant for controller design.
The main question in Method 1 is whether the desired properties of the
continuous-time closed loop system that the designed controller yielded will
be preserved and, if so, in what sense for the closed-loop sampled-data system.
Similarly, the central question in Method 2 is whether or not the properties
of the closed-loop system consisting of the exact discrete-time plant model
and the discrete-time controller will have similar properties as the closed-
loop system consisting of the approximate discrete-time plant model and
the discrete-time controller. In this paper, we present some answers to these
questions for both methods. In particular, we overview several recent results
on Methods 1 and 2 for general nonlinear plants that appeared in [21,27–
31,41]. These results do not contain algorithms for controller design. In the
case of Method 1, controller design is the topic of the area of continuous-time
nonlinear control. In the case of Method 2, controller design algorithms are
yet to be developed and our results provide a unified framework for doing
this.
The paper is organized as follows. In Section 2 we present results on
Method 1 and in Section 3 we present results on Method 3. We do not present
any proofs and the reader may refer to original references for proofs of all
results.

2 Method 1: Emulation

Many results for nonlinear sampled-data systems uses controller emulation

because of the simplicity of the method and the fact that a wide range of
continuous-time controller design methods can be directly used for design of
digital controllers, see [3,33,34,42,41]. An important drawback of the exist-
 Sampled-data control of nonlinear systems 3

ing nonlinear sampled-data theory is that only the questions of stability (see
[3,33,34,42]) and input-to-state stability (see [41]) were addressed within the
emulation design framework. However, one may be interested in a range of
other important system theoretic properties, such as passivity and Lp stabil-
ity. It turns out that a rather general notion of dissipativity that we consider
((V, w)-dissipativity, see Definition 1) can be used to cover a range of most
important system theoretic properties within a unified framework. Special
cases of (V, w)-dissipativity are: stability, input-to-state stability, passivity,
Lp stability, forward completeness, unboundedness observability, etc. Results
in this section are taken from [30] and are concerned with preservation of
(V, w)-dissipativity under sampling in the emulation design for the cases of
static state feedback controllers and open loop controls. Results on emulation
of dynamic state feedback controllers can be found in [21] and they are not
presented here for space reasons. Hence, results in [21,30] provide a general
and unified framework for digital controller design using Method 1.

2.1 Preliminaries
In order to state our results precisely, we need to state some definitions and
assumptions. Sets of real and natural numbers are respectively denoted as
R and N. The Euclidean norm of a vector x is denoted as |x|. Given a set
B ⊂ Rn , its ² neighborhood is denoted as N (B, ²) := {x : inf z∈B |x − z| ≤ ²}.
A function γ : R≥0 → R≥0 is of class-N if it is continuous and non-decreasing.
It is of class-K if it is continuous, zero at zero and strictly increasing; it
is of class-K∞ if it is of class-K and is unbounded. A continuous function
β : R≥0 × R≥0 → R≥0 is of class-KL if β(·, τ ) is of class-K for each τ ≥ 0 and
β(s, ·) is decreasing to zero for each s > 0. For a given function d(·), we use the
following notation d[t1 , t2 ] := {d(t) : t ∈ [t1 , t2 ]}. If t1 = kT, t2 = (k + 1)T , we
use the shorter notation d[k], and take the norm of d[k] to be the supremum
of d(·) over [kT, (k + 1)T ], that is

kd[k]k∞ = ess sup |d(τ )| .

τ ∈[kT,(k+1)T ]

Consider the continuous-time nonlinear plant:

ẋ = f (x, u, dc , ds ), (1)
y = h(x, u, dc , ds ), (2)

where x ∈ Rn , u ∈ Rm and y ∈ Rp are respectively the state, control in-

put and the output of the system and dc ∈ Rnc and ds ∈ Rns are distur-
bance inputs to the system. It is assumed that f and h are locally Lipschitz,
f (0, 0, 0, 0) = 0 and h(0, 0, 0, 0) = 0.
A starting point in the emulation design is to assume that a continuous-
time (open-loop or closed loop system with an appropriately designed continuous-
time controller) possesses a certain property, such as stability. We will assume
4 D. Nešić and A.R. Teel

in the sequel that the continuous system satisfies the following dissipation
property:

Definition 1. The system (1), (2) is said to be (V, w)-dissipative if there

exist a continuously differentiable function V , called the storage function, and
a continuous function w : Rn × Rm × Rnc × Rns → R, called the dissipation
rate, such that for all x ∈ Rn , u ∈ Rm , dc ∈ Rnc , ds ∈ Rns the following holds:
∂V
f (x, u, dc , ds ) ≤ w(x, u, dc , ds ). (3)
∂x
¥

(V, w)-dissipation is a rather general property whose special cases are

stability, input-to-state stability, passivity, etc.
Besides the continuous-time model (1), (2), we consider the exact discrete-
time model for (1), (2) when some of the variables in function f of (1), (2)
are sampled or assumed piecewise constant. More precisely, let T > 0 be a
sampling period and suppose that u and ds in f in (1) are constant during
the sampling intervals, so that u(t) = u(kT ) =: u(k) and ds (t) = ds (kT ) =:
ds (k), ∀t ∈ [kT, (k+1)T ), ∀k ≥ 0, and y is measured only at sampling instants
kT , k ≥ 0.
The exact discrete-time model for the system (1), (2) is obtained by in-
tegrating the initial value problem

ẋ(t) = f (x(t), u(k), dc (t), ds (k)), (4)

with given ds (k), dc [k], u(k) and x0 = x(k), over the sampling interval
[kT, (k + 1)T ]. Let x(t) denotes the solution of the initial value problem (4)
with given ds (k), dc [k], u(k) and x0 = x(k). Then, we can write the exact
discrete-time model for (1), (2) as:
Z (k+1)T
x(k + 1) = x(k) + f (x(τ ), u(k), dc (τ ), ds (k))dτ
kT
:= FTe (x(k), u(k), dc [k], ds (k)) , (5)
y(k) = h(x(k), u(k), dc (k), ds (k)) . (6)

The sampling period T is assumed to be a design parameter which can be

arbitrarily assigned. In practice, the sampling period T is fixed and our results
could be used to determine if it is suitably small. We emphasize that FTe in
(5) is not known in most cases. We denote dc := dc (0) and use it in the sequel.

2.2 Main results

We now present a series of results which provide a general framework for the
emulation design method. In Theorems 1 and 2 we respectively consider the
 Sampled-data control of nonlinear systems 5

“weak” and the “strong” dissipation inequalities for the exact discrete-time
model of the sampled-data system. Each of these dissipation inequalities is
useful in certain situations, as illustrated in the last subsection, where they
are applied to problems of input-to-state stability and passivity. Then several
corollaries are stated for the closed-loop system consisting of (1), (2) and
different kinds of static state feedback controllers. Examples are presented to
illustrate different conditions in main results. For simplicity we do not present
results on emulation of dynamic feedback controllers and these results can
be found in [21].

Theorem 1. (Weak form of dissipation) If the system (1), (2) is (V, w)-
dissipative, then given any 6-tuple of strictly positive real numbers
(∆x , ∆u , ∆dc , ∆d˙c , ∆ds , ν), there exists T ∗ > 0 such that for all T ∈ (0, T ∗ )
and all |x| ≤ ∆x , |u| ≤ ∆u , |ds | ≤ ° ∆ds °and functions dc (t) that are Lipschitz
° °
and satisfy kdc [0]k∞ ≤ ∆dc and °d˙c [0]° ≤ ∆d˙c , the following holds for the
∞
exact discrete-time model (5) of the system (1), (2):

∆V V (FTe (x, u, dc [0], ds )) − V (x)

:= ≤ w(x, u, dc , ds ) + ν . (7)
T T
¥

Under slightly stronger conditions we can prove a stronger result that is

useful in some situations:
Lemma 1. If the system (1), (2) is (V, w)-dissipative, with ∂V ∂x being locally
Lipschitz and ∂V ∂x (0) = 0, then given any quintuple of strictly positive real
numbers (∆x , ∆u , ∆dc , ∆d˙c , ∆ds ), there exist T ∗ > 0 and positive constants
K1 , K2 , K3 , K4 , K5 such that for all T ∈ (0, T ∗ ) and all |x| ≤ ∆x , |u| ≤ ∆u ,
|d
° s | ≤°∆ds and functions dc (t) that are Lipschitz and satisfy kdc [0]k∞ ≤ ∆dc ,
°˙ °
°dc [0]° ≤ ∆d˙c , we have:
∞

∆V
≤ w(x, u, dc , ds ) +
T µ ° ° ¶
2 2 2 2 ° ˙ °2
T K1 |x| + K2 |u| + K3 |ds | + K4 kdc [0]k∞ + K5 °dc [0]° . (8)
∞

In the following theorem we use the strong form of dissipation inequality

for the exact discrete-time model. This result is much more natural to use in
the situations when the disturbances dc are not globally Lipschitz (see the
ISS application in the next subsection and Example 1).

Theorem 2. (Strong form of dissipation) If the system (1), (2) is (V, w)-
dissipative, then given any quintuple of strictly positive real numbers
6 D. Nešić and A.R. Teel

(∆x , ∆u , ∆dc , ∆ds , ν) there exists T ∗ > 0 such that for all T ∈ (0, T ∗ ) and all
|x| ≤ ∆x , |u| ≤ ∆u , kdc [0]k∞ ≤ ∆dc , and |ds | ≤ ∆ds the following holds for
the system (5):
Z
∆V 1 T
≤ w(x, u, dc (τ ), ds )dτ + ν . (9)
T T 0
¥

It is very important to state and prove Theorems 1 and 2 for the case
when a feedback controller is used to achieve (V, w)-dissipativity of the closed
loop system. To prove result on weak dissipation inequalities, we consider the
situation when static state feedback controller of the form:
u = u(x, dc , ds ) (10)
is applied to the system (1), (2). It is assumed below that the feedback (10)
is bounded on compact sets. Note that this general form of feedback covers
both, the full state (u = u(x)) and output (u = u(y)) static feedback. Note
also, that the dissipation rate for the closed-loop system (1), (2) and (10) in
the definition of (V, w)-dissipativity can be taken as w = w(x, dc , ds ). Direct
consequences of Theorem 1 and Lemma 1 are the following corollaries:
Corollary 1. If the system (1), (2), (10) is (V, w)-dissipative, then given
any quintuple of strictly positive real numbers (∆x , ∆dc , ∆d˙c , ∆ds , ν), there
exists T ∗ > 0 such that for all T ∈ (0, T ∗ ) and all |x| ≤ ∆x , |ds |°≤ ∆d°s and
° °
all functions dc (t) that are Lipschitz and satisfy kdc [0]k∞ ≤ ∆dc , °d˙c [0]° ≤
∞
∆d˙c , the following holds for the discrete-time model of the closed-loop system
(1), (2), (10):
∆V
≤ w(x, dc , ds ) + ν . (11)
T
¥

Corollary 2. If the system (1), (2), (10) is (V, w)-dissipative, with ∂V ∂x and
u(x, dc , ds ) in (10) being locally Lipschitz and ∂V ∂x (0) = 0 and u(0, 0, 0) = 0,
then given any quadruple of strictly positive real numbers (∆x , ∆dc , ∆d˙c , ∆ds ),
there exists T ∗ > 0 and positive constants K1 , K2 , K3 , K4 such that for all
T ∈ (0, T ∗ ) and all |x| ≤ ∆x , |ds | ≤° ∆ds °and functions dc (t) that are Lipschitz
° °
and satisfy kdc [0]k∞ ≤ ∆dc and °d˙c [0]° ≤ ∆d˙c , the closed-loop discrete-
∞
time model for system (1), (2) and (10) satisfies:

∆V
≤ w(x, dc , ds ) +
T µ ° ° ¶
2 2 2 ° ˙ °2
T K1 |x| + K2 |ds | + K3 kdc [0]k∞ + K4 °dc [0]° .
∞

¥
 Sampled-data control of nonlinear systems 7

It is interesting to note that the condition on the derivative dc in Theorem

1 is necessary to prove the weak dissipation inequality for the discrete-time
system. The following example illustrates this.

Example 1. Consider the continuous time system:

ẋ = u(x) + dc = −x + dc , (12)

where x, dc ∈ R. Using the storage function V = 21 x2 , the derivative of V

is V̇ = −x2 + xdc ≤ − 12 x2 + 12 d2c , and (12) is ISS. We will show that if
¡ ¢
dc (t) = cos t+2T
T the claim of Theorem 1 does not hold since
° ° ° µ ¶°
° 1 t + 2T °
°˙°
°dc ° = ° − sin ° = 1, (13)
∞ ° T T ° T
∞

which goes to infinity as T → 0. Assume that u(x) in (12) is piecewise

constant for the sampled-data system. So, the discrete-time model of the
sampled-data system is
Z (k+1)T µ ¶
τ + 2T
x(k + 1) = (1 − T )x(k) + cos dτ , (14)
kT T

and hence the exact discrete-time model is given by:

x(k + 1) = (1 − T )x(k) + T [sin(k + 3) − sin(k + 2)] , ∀k ≥ 0. (15)

Suppose that for any given ∆x , ∆dc and ν, there exists T ∗ > 0 such that for
all T ∈ (0, T ∗ ) and k ≥ 0 with |x| ≤ ∆x and kdc [0]k∞ ≤ ∆dc we have

∆V 1 1
≤ − x2 + d2c + ν. (16)
T 2 2
We show by contradiction that the claim is not true for our case. Direct
computations yield:
2
∆V ((1 − T )x + T [sin(3) − sin(2)]) − x2
=
T 2T
= −x2 + x [sin(3) − sin(2)] + O(T ). (17)

Let x̃ = −0.5, (and hence ∆x = 0.5, ∆dc = 1). By combining (16) and (17)
we conclude that there should exist T ∗ > 0 such that ∀T ∈ (0, T ∗ ) we obtain:

1 1
− x̃2 + x̃ [sin(3) − sin(2)] − cos(2)2 − ν + O(T ) ≤ 0 , (18)
2 2
and since there exists ν ∗ > 0 such that − 12 x̃2 + x̃ [sin(3) − sin(2)]− 12 cos(2)2 =
ν ∗ we can rewrite (18) as ν ∗ −ν +O(T ) ≤ 0 , which is a contradiction (it does
8 D. Nešić and A.R. Teel

not hold for ν ∈ (0, ν ∗ )). Therefore, for ∆x = 0.5, ∆dc = 1, ν < ν ∗ , there
exists no T ∗ > 0, such that ∀T ∈ (0, T ∗ ) the condition
° ° (16) holds. Note that
° °
the chosen dc (t) does not satisfy the condition °d˙c ° ≤ ∆d˙c for some fixed
∞
∆d˙c > 0, which is evident from (13). Hence, in this case we can not apply
Theorem 1. N

To state results on strong dissipation inequalities with static state feed-

back controllers, we need to consider controllers of the following form:

u = u(x, ds ) , (19)

in order to be able to state a general results on strong dissipation inequalities.

This is illustrated by the following example.

Example 2. Consider the system ẋ = u, and u = −dc , where dc (0) = 0

and dc (t) = 1, ∀t > 0. Using V (x) = x, such that ∂V ∂x (−dc ) = −dc and
w(x, dc , ds ) = −dc . Since u is sampled and dc (0) = 0, we have that x(t) =
RT
0, ∀t ∈ [0, T ] and so ∆V /T = 0. On the other hand 0 w(dc (τ ))dτ = −T .
Hence, if (20) was true, then we would obtain obtain 0 ≤ −1 + ν, which is
not true for small ν. ¥

Corollary 3. If the system (1), (2), (19) is (V, w)-dissipative, then given
any quadruple of strictly positive real numbers (∆x , ∆dc , ∆ds , ν), there exists
T ∗ > 0 such that for all T ∈ (0, T ∗ ) and all |x| ≤ ∆x , kdc [0]k∞ ≤ ∆dc , and
|ds | ≤ ∆ds the following holds for the closed-loop discrete-time model of the
system (1), (2) and (19):
Z T
∆V 1
≤ w(x, dc (τ ), ds )dτ + ν . (20)
T T 0

Corollary 4. If the system (1), (2), (19) is (V, w)-dissipative, with ∂V ∂x and
u(x, ds ) in (19) being locally Lipschitz and ∂V∂x (0) = 0, u(0, 0) = 0, then
given any triple of strictly positive real numbers (∆x , ∆dc , ∆ds ), there exists
T ∗ > 0 and positive constants K1 , K2 , K3 such that for all T ∈ (0, T ∗ ) and
all |x| ≤ ∆x , kdc [0]k∞ ≤ ∆dc , and |ds | ≤ ∆ds the closed-loop discrete-time
model for the system (1), (2) and (19) satisfies:
Z T ³ ´
∆V 1 2
≤ w(x, dc (τ ), ds )dτ + T K1 |x|2 + K2 |ds |2 + K3 kdc [0]k∞ .
T T 0

¥
 Sampled-data control of nonlinear systems 9

2.3 Applications
The weak and strong discrete-time dissipation inequalities in Theorems 1 and
2 are tools that can be used to show that the trajectories of the sampled-
data system with an emulated controller have a certain property. In order to
illustrate what kind of properties can be proved for the sampled-data system
using the weak or strong inequalities, we apply our results to two impor-
tant system properties: input-to-state stability and passivity. The results on
input-to-state stability were proved in [30] and in [31]. The sketch of proof
of the result on passivity can be found in [30]. Further applications of weak
and strong inequalities to investigation of Lp stability, integral ISS, etc. are
possible and are interesting topics for further research.

Input-to-state stability: Let us suppose that the system

ẋ(t) = f˜(x(t), u(t), dc (t)) (21)

can be rendered ISS using the locally Lipschitz static state feedback

u = u(x), (22)

in the following sense:

Definition 2. The system ẋ = f (x, dc ) is input-to-state stable if there exists
β ∈ KL and γ ∈ K such that the solutions of the system satisfy |x(t)| ≤
β(|x(0)| , t) + γ(kdc k∞ ), ∀x(0), dc ∈ L∞ , ∀t ≥ 0 . ¥

Suppose also that the feedback needs to be implemented using a sampler and
zero order hold, that is:

u(t) = u(x(k)) t ∈ [kT, (k + 1)T ), k ≥ 0 (23)

The following result was first proved in [41] and an alternative proof was
presented in [30]. The proof in [30] is based on the result on strong dissipation
inequalities given in Corollary 3 and the results in [27]. In this case, the results
on weak dissipation inequalities could not be used. This is because we do
not want to impose the condition that the disturbances are Lipschitz when
proving the following result, and that is a standing assumption in results on
weak dissipation inequalities.

Corollary 5. If the continuous time system (21), (22) is ISS, then there
exist β ∈ KL, γ ∈ K such that given any triple of strictly positive numbers
(∆x , ∆dc , ν), there exists T ∗ > 0 such that for all T ∈ (0, T ∗ ), |x(t0 )| ≤ ∆x ,
kdc k∞ ≤ ∆dc , the solutions of the sampled-data system (21), (23) satisfy:

|x(t)| ≤ β(|x(t0 )|, t − t0 ) + γ(kdc k∞ ) + ν, (24)

∀t ≥ t0 ≥ 0. ¥
10 D. Nešić and A.R. Teel

Corollary 5 states that if the continuous-time closed loop system is ISS,

then the sampled-data system with the emulated controller will be semiglob-
ally practically ISS, where the sampling period is the parameter that we can
adjust. Besides the above given property that is presented in an L∞ setting,
we can prove the following integral (or L2 ) version of the same result that
was proved in [31].
Corollary 6. If the system (21), (22) is ISS, then given any quadruple of
strictly positive real numbers (∆x , ∆w , ν1 , ν2 ) there exists T ∗ > 0 such that
for all T ∈ (0, T ∗ ), |x(0)| ≤ ∆x and kwk∞ ≤ ∆w , the following inequality
holds for the sampled-data system (21), (23) satisfy:
Z t Z t
α3 (|x(s)|)ds ≤ α2 (|x(0)|) + γ(|w(s)|)ds + ν1 t + ν2 , (25)
0 0

for all t ≥ 0. ¥

Passivity: Consider the continuous time system with outputs

ẋ = f (x, u), y = h(x, u), (26)
where x ∈ Rn , y, u ∈ Rm and assume that the system is passive, that is
(V, w)-dissipative, where V : Rn → R≥0 and w = y T u. We can apply either
results of Theorem 1 or 2 since u is a piecewise constant input, to obtain
that the discrete-time model satisfies the following: for any (∆x , ∆u , ν) there
exists T ∗ > 0 such that for all T ∈ (0, T ∗ ), |x| ≤ ∆x , |u| ≤ ∆u we have:
∆V
≤ y T u + ν. (27)
T
In stability and ISS applications, adding ν in the dissipation inequality dete-
riorated the property, but the deterioration was gradual. However, in (27) ν
acts as an infinite energy storage (finite power source) and hence it contra-
dicts the definition of a passive system as one that can not generate power
internally. As a result, conditions which guarantee that ν in (27) can be
set to zero are very important. These conditions are spelled out in the next
corollary:
Corollary 7. Suppose that the system (26) is strictly input and state passive
in the following sense: the storage function has gradient ∂V ∂x that is locally
Lipschitz and zero at zero and the dissipation rate can be taken as w(x, y, u) =
y T u − ψ1 (x) − ψ2 (u), where ψ1 and ψ2 are positive definite functions that are
locally quadratic. Then given any pair of strictly positive numbers (∆x , ∆u )
there exists T ∗ > 0 such that for all T ∈ (0, T ∗ ), |x| ≤ ∆x , |u| ≤ ∆u we have:
∆V 1 1
≤ y T u − ψ1 (x) − ψ2 (u). (28)
T 2 2
¥
 Sampled-data control of nonlinear systems 11

3 Method 2: Approximate discrete-time model design

As we already indicated in the introduction, a majority of nonlinear and linear
discrete-time control literature is based on the assumption, which is unreal-
istic in general, that the exact discrete-time model of the plant is known. In
reality, even in the case of linear systems we use an approximate discrete-
time model for the discrete-time controller design. We recognize this fact and
whenever we refer to Method 2, we always assume that only an approximate
discrete-time model is available for the controller design.
One may be tempted to believe that if a controller is designed for an ap-
proximate discrete-time model of the plant with a sufficiently small sampling
period then the same controller will stabilize the exact discrete-time model.
Note that if this was true then one could directly apply all the existing the-
ory that assumes that the exact discrete-time model is known. However, this
reasoning is wrong since no matter how small the sampling period is, we may
always find a controller that stabilizes the approximate model for that sam-
pling period but destabilizes the exact model for the same sampling period,
as illustrated by the following example.
Example 3. Consider the triple integrator ẋ1 = x2 , ẋ2 = x3 , ẋ3 = u, its
Euler approximate model

x1 (k + 1) = x1 (k) + T x2 (k)
x2 (k + 1) = x2 (k) + T x3 (k) (29)
x3 (k + 1) = x3 (k) + T u(k) ,

and a minimum-time dead-beat controller for the Euler discrete-time model

given by
µ ¶
x1 (k) 3x2 (k) 3x3 (k)
u(k) = − 3 − − . (30)
T T2 T
The closed loop system consisting of (29) and (30) has all poles equal to zero
for all T > 0 and hence this discrete-time Euler-based closed loop system
is asymptotically stable for all T > 0. On the other hand, the closed loop
system consisting of the exact discrete-time model of the triple integrator
and controller (30) has a pole at ≈ −2.644 for all T > 0. Hence, the closed-
loop sampled-data control system is unstable for all T > 0 (and, hence, also
for arbitrarily small T )! So we see that, to design a stabilizing controller
using Method 2, it is not sufficient to design a stabilizer for an approximate
discrete-time model of the plant for sufficiently small T . Extra conditions are
needed!
Several control laws in the literature have been designed based on ap-
proximate discrete-time models of the plant, see [8,12,23]. These results are
always concerned with a particular plant model and a particular approxi-
mate discrete-time model (usually the Euler approximation) and hence they
12 D. Nešić and A.R. Teel

are not very general. On the other hand, we present a rather general result
for a large class of plants, a large class of approximate discrete-time models
and the conditions we obtain are readily checkable. For different approximate
discretization procedures see [25,38,39,37].
Results in this section are based mainly on [28] and they address the
design of stabilizing static state feedback controllers based on approximate
discrete-time plant models. A more general result was recently proved in
[29] where conditions are presented for dynamic feedback controllers that are
designed for approximate discrete-time models of sampled-data differential
inclusions to prove stability with respect to arbitrary non-compact sets.

3.1 Main results

Consider the nonlinear continuous-time control system:

ẋ = f (x, u) x(0) = x◦ (31)

where x ∈ Rn , u ∈ Rm are respectively the state and control vectors. The

function f is assumed to be such that, for each initial condition and each
constant control, there exists a unique solution defined on some (perhaps
bounded) interval of the form [0, τ ). The control is taken to be a piecewise
constant signal u(t) = u(kT ) =: u(k), ∀t ∈ [kT, (k + 1)T [, k ∈ N where
T > 0 is a sampling period. We assume that the state measurements x(k) :=
x(kT ) are available at sampling instants kT, k ∈ N. The sampling period T
is assumed to be a design parameter which can be arbitrarily assigned (in
practice, the sampling period T is fixed and our results could be used to
determine if it is suitably small).
Suppose that we want to design a control law for the plant (31) using
Method 2. The controller will be implemented digitally using a sampler and
zero order hold element. As a first step in this direction we need to specify a
discrete-time model of the plant (31), which describes the behavior of the sys-
tem at sampling instants. We consider the difference equations corresponding
to the exact plant model and its approximation respectively:
x(k + 1) = FTe (x(k), u(k)) (32)
x(k + 1) = FTa (x(k), u(k)) (33)
which are parameterized with the sampling period T . We emphasize that
FTe is not known in most cases. We will think of FTe and FTa as being defined
globally for all small T even though the initial value problem (31) may exhibit
finite escape times. We do this by defining FTe arbitrarily for pairs (x(k), u(k))
corresponding to finite escapes and noting that such points correspond only
to points of arbitrarily large norm as T → 0, at least when f is locally
bounded. So, the behavior of FTe will reflect the behavior of (31) as long as
(x(k), uT (x(k))) remains bounded with a bound that is allowed to grow as
T → 0. This is consistent with our main results that guarantee practical
 Sampled-data control of nonlinear systems 13

asymptotic stability that is semiglobal in the sampling period, i.e., as T →

0 the set of points from which convergence to an arbitrarily small ball is
guaranteed to contain an arbitrarily large neighborhood of the origin.
In general, one needs to use small sampling periods T since the approxi-
mate plant model is a good approximation of the exact model typically only
for small T . It is clear then that we need to be able to obtain a controller
uT (x) which is, in general, parameterized by T and which is defined for all
small T . For simplicity, we consider only static state feedback controllers.
For a fixed T > 0, consider systems (32), (33) with a given controller
u(k) = uT (x(k)). We denote the state of the closed-loop system (32) (respec-
tively (33)) with the given controller at time step k that starts from x(0) as
xe (k, x(0)) or xek (respectively xa (k, x(0)) or xak ). We introduce the error:

εk (ξ, z) := xe (k, ξ) − xa (k, z), (34)

and also use the notation εk (ξ) := εk (ξ, ξ) = xe (k, ξ) − xa (k, ξ). In our
results (see Theorems 3 and 4), we will make a stability assumption on the
family of closed-loop approximate plant models and will draw a conclusion
about stability of the family of closed-loop exact plant models by invoking
assumptions about the closeness of solutions between the two families.

One-step consistency The first type of closeness we will use is character-

ized in the following definition. It guarantees that the error between solutions
starting from the same initial condition is small, over one step, relative to
the size of the step. The terminology we use is based on that used in the
numerical analysis literature (see [39]).

Definition 3. The family (uT , FTa ) is said to be one-step consistent with

(uT , FTe ) if, for each compact set X ⊂ Rn , there exist a function ρ ∈ K∞ and
T ∗ > 0 such that, for all x ∈ X and T ∈]0, T ∗ [, we have

|FTe (x, uT (x)) − FTa (x, uT (x))| ≤ T ρ(T ) . (35)

A sufficient condition for one-step consistency is the following:

Lemma 2. If

1. (uT , FTa ) is one-step consistent with (uT , FTEuler ) where FTEuler (x, u) :=
x + T f (x, u),
2. for each compact set X ⊂ Rn there exist ρ ∈ K∞ , M > 0, T ∗ > 0 such
that, for all T ∈]0, T ∗ [ and all x, y ∈ X ,
(a) |f (y, uT (x))| ≤ M ,
(b) |f (y, uT (x)) − f (x, uT (x))| ≤ ρ(|y − x|),

then (uT , FTa ) is one-step consistent with (uT , FTe ).

14 D. Nešić and A.R. Teel

Multi-step consistency The second type of closeness we will use is char-

acterized in terms of the functions FTe , FTa and uT (x) in the next definition.
It will guarantee (see Lemma 3) that the error between solutions starting
from the same initial condition is small over multiple steps corresponding
“continuous-time” intervals with length of order one.
Definition 4. The family (uT , FTa ) is said to be multi-step consistent with
(uT , FTe ) if, for each L > 0, η > 0 and each compact set X ⊂ Rn , there exist a
function α : R≥0 ×R≥0 → R≥0 ∪ {∞} and T ∗ > 0 such that, for all T ∈]0, T ∗ [
we have that {x, z ∈ X , |x − z| ≤ δ} implies

|FTe (x, uT (x)) − FTa (z, uT (z))| ≤ α(δ, T ) (36)

and
k
z }| {
k
k ≤ L/T =⇒ α (0, T ) := α (· · · α (α (0, T ) , T ) · · · , T ) ≤ η .
(37)

In terms of trajectory error over “continuous-time” intervals with length of

order one, multi-step consistency gives the following:
Lemma 3. If (uT , FTa ) is multi-step consistent with (uT , FTe ) then for each
compact set X ⊂ Rn , L > 0 and η > 0 there exists T̂ > 0 such that, if T and
ξ satisfy

T ∈]0, T̂ [ , xak (ξ) ∈ X ∀k : kT ∈ [0, L], (38)

then

|εk (ξ)| ≤ η ∀k : kT ∈ [0, L] . (39)

An interesting sufficient condition for multi-step consistency is given in

the following:
Lemma 4. If, for each compact set X ⊂ Rn , there exist K > 0, ρ ∈ K∞ and
T ∗ > 0 such that for all T ∈]0, T ∗ [ and all x, z ∈ X we have

|FTe (x, uT (x)) − FTa (z, uT (z))| ≤ (1 + KT ) |x − z| + T ρ(T ) (40)

then (uT , FTa ) is multi-step consistent with (uT , FTe ).

Relative to the one-step consistency condition, the condition of Lemma
4 is guaranteed by one-step consistency plus the following type of Lipschitz
condition on either the family (uT , FTe ) or the family (uT , FTa ):
for each compact set X ⊂ Rn there exist K > 0 and T ∗ > 0 such that for all
x, z ∈ X and all T ∈]0, T ∗ [,

|FT (x, uT (x)) − FT (z, uT (z))| ≤ (1 + KT )|x − z| . (41)

 Sampled-data control of nonlinear systems 15

This condition is guaranteed for FTe when f (x, u) and uT (x) are locally Lip-
schitz (uniformly in small T ). Note that no continuity assumptions on uT (x)
were made in Lemma 2 to guarantee one-step consistency. The condition
given in Lemma 4 for multi-step consistency is similar to conditions used in
the numerical analysis literature (e.g., see conditions (i) and (iii) of Assump-
tion 6.1.2 in [39, pg.429]).
One-step and multi-step consistency do not imply each other and this is
one motivation for developing different stability theorems that rely on one of
these properties. Example 4 in Section 3.3 shows that multi-step consistency
may not hold when one-step consistency does hold. That one-step consistency
may not hold when multi-step consistency does hold can be seen from the
plant ẋ = x + u with Euler approximation x(k + 1) = x(k) + T (x(k) + u(k)) =
FTa (x(k), u(k)) and controller uT (x) = −( T1 + 1)x. The exact discrete-time
model is x(k + 1)³ = eT x(k)´+ (eT − 1)u(k) and we have FTa (x, uT (x)) ≡ 0 and
T
FTe (x, uT (x)) = 1 − e T−1 x. Since, for x in a compact set, FTe (x, uT (x)) is
of order T we do not have one-step consistency. On the other hand, it follows
from FTa (x, uT (x)) ≡ 0 and the fact that FTe (x, uT (x)) is of order T that we
do have multi-step consistency. Indeed, for each compact set X ⊂ R and each
η > 0 there exist strictly positive numbers K, T ∗ such that, for all x, z ∈ X ,
T ∈]0, T ∗ [, k ≥ 0,

|FTe (x, uT (x)) − FTa (z, uT (z))| = |FTe (x, uT (x))| ≤ KT

:= α(δ, T ) = αk (0, T ) ≤ η .

3.2 Stability properties

We now give conditions on the family (uT , FTa ) that guarantee asymptotic
stability for the family (uT , FTe ). As we have already seen in Example 3, it
is not enough to assume simply that each member of the family (uT , FTa ) is
asymptotically stable (at least for small T ). Instead, we will impose partial
uniformity of the stability property over all small T . For that, we make the
following definitions:

Definition 5. Let β ∈ KL and let N ⊂ Rn be an open (not necessarily

bounded) set containing the origin.

1. The family (uT , FT ) is said to be (β, N )-stable if there exists T ∗ > 0 such
that for each T ∈]0, T ∗ [, the solutions of the system

x(k + 1) = FT (x(k), uT (x(k))) (42)

satisfy

|x(k, x(0))| ≤ β(|x(0)| , kT ), ∀x(0) ∈ N, k ≥ 0. (43)

16 D. Nešić and A.R. Teel

2. The family (uT , FT ) is said to be (β, N )-practically stable if for each

R > 0 there exists T ∗ > 0 such that for each T ∈]0, T ∗ [ the solutions of
(42) satisfy:
|x(k, x(0))| ≤ β(|x(0)| , kT ) + R, ∀x(0) ∈ N, k ≥ 0. (44)
An equivalent Lyapunov formulation of (β, Rn )-stability is the following
(local versions can also be formulated but are more tedious to state because
of the need to keep track of basins of attraction):
Lemma 5. The following statements are equivalent:
1. There exists β ∈ KL such that the family (uT , FT ) is (β, Rn )-stable.
2. There exist T ∗ > 0, α1 , α2 ∈ K∞ , α3 ∈ K and for each T ∈]0, T ∗ [, VT :
Rn → R≥0 such that ∀x ∈ Rn , ∀T ∈]0, T ∗ [ we have:
α1 (|x|) ≤ VT (x) ≤ α2 (|x|), (45)
VT (FT (x, uT (x))) − VT (x) ≤ −T α3 (|x|), . (46)
In our first main result (Theorem 3) we will show that if the family
(uT , FTa ) is (β, N )-stable and multi-step consistent with (uT , FTe ) then the
family (uT , FTe ) is (β, N )-practically stable. We will also show (in Theorem 4)
that the multi-step consistency assumption can be changed to a one-step con-
sistency assumption when (β, N )-stability is formulated in terms of a family
of Lyapunov functions satisfying (45),(46) and with an extra local Lipschitz
condition that is uniform in small T .
Definition 6. The family (uT , FT ) is said to be equi-globally asymptotically
stable (EGAS) by equi-Lipschitz Lyapunov functions if the second statement
of Lemma 5 holds and, moreover, for each compact set X ⊂ Rn \ {0} there
exist M > 0 and T ∗ > 0 such that, for all x, z ∈ X and all T ∈]0, T ∗ [,
|VT (x) − VT (z)| ≤ M |x − z| . (47)
Our first result is expressed in terms of trajectory bounds for (uT , FTa )
and multi-step consistency:
Theorem 3. Let β ∈ KL and let N be a bounded neighborhood of the origin.
If the family (uT , FTa ) is:
A: multi-step consistent with (uT , FTe ), and
B: (β, N )-stable,
then
C: the family (uT , FTe ) is (β, N )-practically stable.
Our second result is expressed in terms of a family of Lyapunov functions
for (uT , FTa ) and one-step consistency. It has some relations to the proof
technique used to establish the main result of [7]. For simplicity we will only
formulate the global result. Nonglobal results and results for stability of sets
other than the origin can also be established with the same proof technique.
 Sampled-data control of nonlinear systems 17

Theorem 4. If
A1: (uT , FTa ) is one-step consistent with (uT , FTe ), and
B1: (uT , FTa ) is EGAS by equi-Lipschitz Lyapunov functions
then
C1: there exists β ∈ KL such that, for each bounded neighborhood N of the
origin, the family (uT , FTe ) is (β, N )-practically stable.

3.3 Examples
Example 4 illustrates Theorem 4 by giving an example where multi-step
consistency does not hold but one-step consistency does hold and there is
a suitable family of Lyapunov functions. Example 5 shows situation where
each element of the family (uT , FTa ) is globally exponentially stable with
overshoots uniform in T but where the family (uT , FTe ) fails to be (β, N )-
practically stable for any pair (β, N ). In reference to Theorem 3, we use the
notation C for this situation.

Example 4. Consider the two-input linear system

ẋ1 = x1 + u1
(48)
ẋ2 = u2

which has exact discretization

x1 (k + 1) = eT x1 (k) + [eT − 1]u1 (k)
(49)
x2 (k + 1) = x2 (k) + T u2 (k)

and Euler approximate discretization

x1 (k + 1) = [1 + T ]x1 (k) + T u1 (k)

(50)
x2 (k + 1) = x2 (k) + T u2 (k) .

Consider the controller

 

 −2x


1
 if 0 < 0.1x1 < x2 < 10x1


 0
u(x) =   (51)

 −2x

 1
 otherwise.



−x2

A1,A: It follows from Lemma 2 that (uT , FTa ) is one-step consistent

with (uT , FTe ). However, (uT , FTa ) is not multi-step consistent with (uT , FTe ).
Indeed, consider the initial condition (ξ1 , ξ2 ) = (1, 0.1). It is easy to see
that, in this case, (xa1 (k, ξ), xa2 (k, ξ)) = (1 − T )k (1, 0.1), i.e., the positive
18 D. Nešić and A.R. Teel

ray x2 = 0.1x1 > 0 is forward invariant for all T ∈ (0, 1). On the other
hand, (xe1 (1, ξ), xe2 (1, ξ)) = ((2 − eT )1, (1 − T )0.1), i.e., for all small T > 0,
xe2 (1, ξ) < 10xe1 (1, ξ) and xe2 (1, ξ) > 0.1xe1 (1, ξ) since eT > 1 + T . It fol-
lows that, for k ≥ 1, x(k, ξ) will take values on the horizontal line given by
x2 = (1 − T )0.1 moving in the direction of decreasing x1 until it crosses the
positive ray x2 = 10x1 . Let k̄ denote the number of steps required to cross the
positive ray x2 = 10x1 . It is easy to put an upper and lower bound on k̄T that
is independent of T . Then since, for all k ≤ k̄, we have xe2 (k, ξ) = (1 − T )0.1
while xa2 (k, ξ) = (1 − T )k 0.1 ≤ e−kT 0.1, it is clear that the conclusion of
Lemma 3 is not satisfied. Hence (uT , FTa ) cannot be multi-step consistent
with (uT , FTe ).
B1: We take VT (x) = |x1 | + |x2 |. We get, for T ∈ (0, 1) and 0 < 0.1x1 <
x2 < 10x1 :
T
VT (FTa (x, uT (x))) − V (x) = −T |x1 | ≤ − [|x1 | + |x2 |] (52)
20
and, otherwise,

VT (FTa (x, uT (x))) − V (x) = −T [|x1 | + |x2 |] . (53)

It follows that the family (uT , FTa ) is EGAS by equi-Lipschitz Lyapunov func-
tions.
C1: We conclude from Theorem 4 (and also using the homogeneity of
VT (x) and FTe (x, uT (x)) to pass from a semiglobal practical result to a global
result and following the steps of the proof of Theorem 4 to get an exponential
result) that the family (uT , FTe ) is (β, R2 )-stable with β(s, t) of the form
ks exp(−λt) with k > 0 and λ > 0.

Example 5. (A, B, C) Consider the double integrator, its Euler approxima-

tion and its exact discrete-time model:
double integrator: ẋ1 = x2
ẋ2 = u (54)
approximate: x1 (k + 1) = x1 (k) + T x2 (k)
x2 (k + 1) = x2 (k) + T u(k) (55)
exact: x1 (k + 1) = x1 (k) + T x2 (k) + 0.5T 2 u(k)
x2 (k + 1) = x2 (k) + T u(k). (56)
The following controller is designed for the Euler model:
x1 2x2
u(x) = − − . (57)
T T
C : The eigenvalues of the exact closed-loop are λ1 = 1− T2 , λ2 = −1, ∀T >
0 and thus the exact closed-loop model is not (β, N )-practically stable for any
pair (β, N ).
 Sampled-data control of nonlinear systems 19
√
√A : The eigenvalues of the Euler closed-loop system are λ1 = + 1 − T , λ2 =
− 1 − T . In a similar way as in the previous example we can show that there
exists b > 0 such that for all T ∈]0, 0.5[ we have:

|x(k)| ≤ b exp(−0.5kT ) |x(0)| , ∀x(0) ∈ R2 .

Hence, the approximate closed-loop system is (β, R2 )-stable with β(s, t) :=

b exp(−0.5t).
B : It now follows from Theorem 3 that (uT , FTa ) is not multi-step consis-
tent with (uT , FTe ). In fact, (uT , FTa ) is not one-step consistent with (uT , FTe )
since
¯ ¯
|²1 (x)| = ¯T 2 /2(−x1 /T − 2x2 /T )¯ = 2T |x1 + x2 | , ∀x ∈ R2 , ∀T.

4 Conclusion
Several recent results on design of sampled-data controllers that appeared in
[21,27–31] were overviewed. These results are geared toward providing a uni-
fied framework for the digital controller design based either on the continuous-
time plant model (Method 1) or on an approximate discrete-time plant model
(Method 2). The conditions we presented are easily checkable and the results
are applicable to a wide range of plants, controllers and system theoretic
properties. Further research is needed to provide control design algorithms
based on approximate discrete-time models. Our results on Method 2 provide
a unified framework for doing so.

References
1. J. P. Barbot, S. Monaco, D. Normand-Cyrot and N. Pantalos, Discretization
schemes for nonlinear singularly perturbed systems, Proc. CDC’91, pp. 443-448.
2. C. I. Byrnes and W. Lin, “Losslessness, feedback equivalence and the global
stabilization of discrete-time systems”, IEEE Trans. Automat. Contr., 39 (1994),
pp. 83-97.
3. B. Castillo, S. Di Gennaro, S. Monaco and D. Normand-Cyrot, On regulation
under sampling, IEEE Trans. Automat. Contr., 42 (1997), pp. 864-868.
4. P. D. Christofides and A. R. Teel, Singular perturbations and input-to-state sta-
bility, IEEE Trans. Automat. Contr., 41 (1996), pp. 1645-1650.
5. T. Chen and B. A. Francis, Input-output stability of sampled-data systems, IEEE
Trans. Automat. Contr., 36 (1991), pp. 50-58.
6. T. Chen and B. A. Francis, Optimal sampled-data control systems. Springer-
Verlag: London, 1995.
7. F. H. Clarke, Y. S. Ledyaev, E. D. Sontag and A. I. Subbotin, Asymptotic
controllability implies feedback stabilization, IEEE Trans. Automat. Contr., 42
(1997), pp. 1394-1407.
8. D. Dochain and G. Bastin, Adaptive identification and control algorithms for
nonlinear bacterial growth systems, Automatica, 20 (1984), pp. 621-634.
20 D. Nešić and A.R. Teel

9. G. F. Franklin, J. D. Powell and M. L. Workman, Digital control of dynamic

systems. Addison-Wesley Pub. Co. Inc.: Reading, 1990.
10. B. A. Francis and T. T. Georgiou, Stability theory for linear time-invariant
plants with periodic digital controllers, IEEE Trans. Automat. Contr., vol. 33
(1988), pp. 820-832.
11. S. T. Glad, Output dead-beat control for nonlinear systems with one zero at
infinity, Systems and Control Letters, 9 (1987), pp. 249-255.
12. G. C. Goodwin, B. McInnis and R. S. Long, Adaptive control algorithm for
waste water treatment ad pH neutralization, Optimal Contr. Applic. Meth., 3
(1982), pp. 443-459.
13. L. Grüne, Input-to-state stability of exponentially stabilized semi-linear control
systems with inhomogeneous perturbations, preprint (1998).
14. L. Hou, A. N. Michel and H. Ye, Some qualitative properties of sampled-data
control systems, IEEE Trans. Automat. Contr., 42 (1997), pp. 1721-1725.
15. P. Iglesias, Input-output stability of sampled-data linear time-varying systems,
IEEE Trans. Automat. Contr., 40 (1995), pp. 1646-1650.
16. A. Iserles and G. Söderlind, Global bounds on numerical error for ordinary
differential equations, J. Complexity, 9 (1993), pp. 97-112.
17. N. Kazatzis and C. Kravaris, System-theoretic properties of sampled-data rep-
resentations of nonlinear systems obtained via Taylor-Lie series, Int. J. Control,
67 (1997), pp. 997-1020.
18. H. K. Khalil, Nonlinear systems. Prentice-Hall: New Jersey, 1996.
19. P.E. Kloeden and J. Lorenz. Stable attracting sets in dynamical systems and
their one-step discretizations, SIAM J. Num. Anal., 23 (1986), pp. 986-995.
20. B. C. Kuo, Digital control systems. Saunders College Publishing: Ft. Worth,
1992.
21. D. S. Laila and D. Nešić, A note on preservation of dissipation inequalities
under sampling: the dynamic feedback case, submitted to Amer. Contr. Conf.,
2001.
22. V. Lakshmikantham and S. Leela, Differential and integral inequalities, vol. 1.
Academic Press: New York, 1969.
23. I. M. Y. Mareels, H. B. Penfold and R. J. Evans, Controlling nonlinear time-
varying systems via Euler approximations, Automatica, 28 (1992), pp. 681-696.
24. S. Monaco and D. Normand-Cyrot, Zero dynamics of sampled nonlinear sys-
tems, Syst. Contr. Lett., 11 (1988), pp. 229-234.
25. S. Monaco and D. Normand-Cyrot, Sampling of a linear analytic control system,
Proc. CDC’85, pp. 1457-1462.
26. M. S. Mousa, R. K. Miller and A. N. Michel, Stability analysis of hybrid compos-
ite dynamical systems: descriptions involving operators and difference equations,
IEEE Trans. Automat. Contr., 31 (1986), pp. 603-615.
27. D. Nešić, A. R. Teel and E.D.Sontag, Formulas relating KL stability estimates of
discrete-time and sampled-data nonlinear systems, Sys. Contr. Lett., 38 (1999),
pp. 49-60.
28. D. Nešić, A. R. Teel and P. V Kokotović, Sufficient conditions for stabilization
of sampled-data nonlinear systems via discrete-time approximations, Syst. Contr.
Lett., 38 (1999), pp. 259-270.
29. D. Nešić and A. R. Teel, Set stabilization of sampled-data differential inclusions
via their approximate discrete-time models, to appear in Conf. Decis. Contr.,
Sydney, 2000.
 Sampled-data control of nonlinear systems 21

30. D. Nešić, D. S. Laila and A. R. Teel, On preservation of dissipation inequalities

under sampling, to appear in Conf. Decis. Contr., Sydney, 2000.
31. D. Nešić and P. Dower, Further results on preservation of input-to-state stability
under sampling, to appear in ICARV, Singapore, 2000.
32. R. Ortega and D. Taoutaou, A globally stable discrete-time controller for
current-fed induction motors, Systems and Control Letters, 28 (1996), pp. 123-
128.
33. D. H. Owens, Y. Zheng and S. A. Billings, Fast sampling and stability of non-
linear sampled-data systems: Part 1. Existence theorems, IMA J. Math. Contr.
Informat., 7 (1990), pp. 1-11.
34. Z. Qu, Robust control of nonlinear uncertain systems. John Wiley & Sons: New
York, 1998.
35. E. D. Sontag, Smooth stabilization implies coprime factorization, IEEE Trans.
Automat. Contr., 34 (1989), 435-443.
36. E. D. Sontag, Remarks on stabilization and input-to-state stability, in Proc.
CDC, Tampa, USA, 1989.
37. S. A. Svoronos, D. Papageorgiou and C. Tsiligiannis, Discretization of nonlin-
ear control systems via the Carleman linearization, Chemical Eng. Science, 49
(1994), pp. 3263-3267.
38. H. J. Stetter, Analysis of discretization methods for ordinary differential equa-
tions. Springer-Verlag: New York, 1973.
39. A. M. Stuart and A. R. Humphries, Dynamical systems and numerical analysis.
Cambridge University Press: New York, 1996.
40. A. R. Teel, Connection between Razumikhin-type theorems and the ISS nonlin-
ear small gain theorem, IEEE Trans. Automat. Contr., to appear, 1998.
41. A. R. Teel, D. Nešić and P. V. Kokotović, A note on input-to-state stability
of sampled-data nonlinear systems, In Proceedings of the 37th IEEE Conference
on Decision and Control, Tampa, Florida, 1998.
42. Y. Zheng, D. H. Owens and S. A. Billings, Fast sampling and stability of non-
linear sampled-data systems: Part 2: Sampling rate estimation, IMA J. Math.
Contr. Informat., 7 (1990), pp. 13-33.